Nvidia BigK GK110 Kepler Speculation Thread

The NVIDIA Titan card will apparently be the replacement for the GeForce GTX 690, a dual-GK104 card launched in May of last year. The performance estimate for the Titan is approximately 85% of that GTX 690

85% of the 690, not over it.
 
Before we have another rewriting of history, it's probably a good time to point out that of the 12 months last year, AMD held the performance crown for 9 of them (and still does).

You know what I meant, no reason to get jumpy.
Had the 7970 been stronger initially, the 680 would have never had the performance crown at all. And Nvidia would have known that full well and lived with it - which would have been a first in a long time.

And no, not 9 months, but 6 until the GHz Edition came out in June. Before that, the 680 was slightly faster, but it was so close that I would also call it a tie.
 
Only at lower resolutions/AA levels, if you piled on the pixels the 680 dropped back due bandwidth starvation (most likely.)

No:
http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_680/27.html

If you look at certain games only, you are of course right. But considering a broad range of games, the 680 and the 7970 were not really apart. +/- up to 5% I would still consider a tie since that can easily change with the selected games.
 
And no, not 9 months, but 6 until the GHz Edition came out in June. Before that, the 680 was slightly faster, but it was so close that I would also call it a tie.

Yes but I meant the 3 months at the start of the year until the 680 was released, then the months since the release of the GHz edition.

AMD led from January to the end of March, or 3 months. Then Nvidia led until the end of June, or 3 months. Since then the GHz edition has been fastest, so it's been AMD for 9 months of last year.

Had the 7970 been stronger initially, the 680 would have never had the performance crown at all. And Nvidia would have known that full well and lived with it - which would have been a first in a long time.
You could also argue that if AMD's drivers had been working properly out of the box instead of waiting until 12.7, then they would have been in a much stronger position too. They made a lot of mistakes last year that they cannot afford to repeat, unfortunately for them it appears we will be back to them being 2nd best for 2013, forced to make do with what they can.
 
85% of the 690, not over it.

That puts it at around 50-60% faster than the 680. That's a big increase for a single GPU generation, definately more than enough to give the new GPU a very clear performance lead if it launches first.

Fingers crossed it's true!
 
That puts it at around 50-60% faster than the 680. That's a big increase for a single GPU generation, definately more than enough to give the new GPU a very clear performance lead if it launches first.

Fingers crossed it's true!

More like ~40%, going by Hardware.fr's numbers.

Still a pretty solid lead over the 7970 GHz Edition, though.
 
If we just assume AMD added more shaders and didn't go crazy (512 bit bus for example), then we should reasonably expect a strong AMD card. I think we are all expecting it to be slower than Nvidia's top card, and we're probably looking at a return to the 580 vs 6970 days.
 
If we just assume AMD added more shaders and didn't go crazy (512 bit bus for example), then we should reasonably expect a strong AMD card.
Don't want to divert too much from thread topic, but AMD only adding shaders would do pretty much nothing at all for gaming performance (which is why the future top chip is still a mystery to me). In any case though I agree it probably should be slower than gk110.
 
Don't want to divert too much from thread topic, but AMD only adding shaders would do pretty much nothing at all for gaming performance (which is why the future top chip is still a mystery to me).
Are you sure about that? At least relative to Nvidia, it seems to me that more shaders would be an improvement. Whereas with GK104, bolting on additional shaders would be nearly pointless. This is well evidenced by the marginal improvement that the 680 offers over the 670.
 
Don't want to divert too much from thread topic, but AMD only adding shaders would do pretty much nothing at all for gaming performance (which is why the future top chip is still a mystery to me). In any case though I agree it probably should be slower than gk110.

Well yes you're correct, I should clarify that I'm assuming AMD has learned somewhat from their underclocked 7-series fiasco and won't make that mistake again (this is far from guaranteed).

We are used to AMD fucking it up, but lets just assume that they aren't completely devoid of sense here. I expect a similar case to the 6970 which was pushed to the limits on clocks and just falling short on performance vs the 580. It might even look good overall when compared to Nvidia's monster die, even if it "loses" in raw performance.
 
Are you sure about that? At least relative to Nvidia, it seems to me that more shaders would be an improvement. Whereas with GK104, bolting on additional shaders would be nearly pointless. This is well evidenced by the marginal improvement that the 680 offers over the 670.

More shaders will help AMD more on 7-series architectures. However Nvidia's step up to 384-bit will probably yield more. I'm honestly not sure, it's a tough call. I suspect Nvidia's throwing more area at the problem will eventually overcome AMD's more "elegant" solution. It'll be 580 vs 6790 all over again or I'll eat my proverbial hat.
 
Problem I see is that AMD needs more shaders and bandwidth to do the same job as Nvidia. Just look at reviews of the 7870 LE. It has basically the same specs as a GTX670 and even 15% more compute power. And yet the 670 is 5-15% faster depending on the review and settings.
Especially in the games and settings where AMD is usually very strong due to higher compute power and bandwidth, the 7870 LE falls back to 670/680 levels or even below. So if Nvidia matches or even surpasses AMD in raw power, AMD will have a real problem on their hands.
 
Problem I see is that AMD needs more shaders and bandwidth to do the same job as Nvidia. Just look at reviews of the 7870 LE. It has basically the same specs as a GTX670 and even 15% more compute power. And yet the 670 is 5-15% faster depending on the review and settings.
Especially in the games and settings where AMD is usually very strong due to higher compute power and bandwidth, the 7870 LE falls back to 670/680 levels or even below. So if Nvidia matches or even surpasses AMD in raw power, AMD will have a real problem on their hands.
That's a good point. It'll be interesting to see what the second iteration of GCN looks like.
 
Back
Top